CN105490959B - Implementation method is embedded in based on the non-homogeneous bandwidth virtual data center that congestion is evaded - Google Patents

Implementation method is embedded in based on the non-homogeneous bandwidth virtual data center that congestion is evaded Download PDF

Info

Publication number
CN105490959B
CN105490959B CN201510932813.9A CN201510932813A CN105490959B CN 105490959 B CN105490959 B CN 105490959B CN 201510932813 A CN201510932813 A CN 201510932813A CN 105490959 B CN105490959 B CN 105490959B
Authority
CN
China
Prior art keywords
server
link
congestion
virtual machine
bandwidth
Prior art date
Application number
CN201510932813.9A
Other languages
Chinese (zh)
Other versions
CN105490959A (en
Inventor
闫芳芳
李�东
胡卫生
Original Assignee
上海交通大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海交通大学 filed Critical 上海交通大学
Priority to CN201510932813.9A priority Critical patent/CN105490959B/en
Publication of CN105490959A publication Critical patent/CN105490959A/en
Application granted granted Critical
Publication of CN105490959B publication Critical patent/CN105490959B/en

Links

Abstract

A kind of non-homogeneous bandwidth virtual data center insertion implementation method evaded based on congestion, the sequence that VM is successively decreased by bandwidth demand are sorted, are first sequentially placed in server with adaptation search method for the first time;Start perturbation procedures when adaptation search method can not place the VM for the first time, it is targeting with the most congestion link of physical network, search contributes maximum bottleneck server to this link load, preferentially will re-start the sequence and placement after the smallest VM unloading of bandwidth required in bottleneck server.The present invention solves the problems, such as the placement of routing issue and virtual machine in virtual data center (VDC) imbedding problem, can obtain VDC insertion success rate more higher than the prior art.

Description

Implementation method is embedded in based on the non-homogeneous bandwidth virtual data center that congestion is evaded

Technical field

The present invention relates to the technology in system for cloud computing field, specifically a kind of non-homogeneous bandwidth evaded based on congestion Virtual data center is embedded in implementation method.

Background technique

With popularizing for cloud computing, data center network (DCN) virtualization technology is attracted extensive attention.Virtual machine technology To dispose multiple virtual machines (VM) example in physical server, these virtual machines pass through a shared physics Network is communicated.When the virtual machine of multiple tenants is when shared bottom data central site network generates competition conflict, due to network Bandwidth, which does not ensure, causes unpredictable communication delay and loss of data, eventually leads to hiring cost raising, and bottom-layer network mentions For the income decline of quotient.Predictable network performance can be by providing virtual data center (VDC) Lai Shixian to tenant. Ballani et al. exists " Towards predictable datacenter networks, " (ACM SIGCOMM Computer Communication Review, vol.41, no.4, pp.242-2532011) in propose and a kind of proposed based on Hose model The abstractdesription of VDC request.In Hose model, virtual data center is the set of the virtual machine of a N platform automorphis, They are communicated by virtual link.

R.Matthias et al. exists " Beyond the stars:Revisiting virtual cluster Embeddings, " (ACM SIGCOMM Computer Communication Review, vol.45, no.3, pp.12-18, 2015.) then propose in: resource may be performed Star topology insertion and be wasted, and being based on thus can branch Hose routing proposition HVC-ACE heuristic algorithm, but it is only applicable to uniform bandwidth request.

Even if having determined the placement location of virtual machine, the Multi-path route distribution based on the description of Hose model is very tired Difficult, the Multi-path route distribution being similarly under Virtual Private Network (VPN).Kodialam et al. " Maximum- Throughput routing of traffic in the hose model. " (U.S.Patent No.7,558, In 209.7Jul.2009.) it is similar in VPN by applying the principle of duality of linear programming to solve the problems, such as.

After searching and discovering the prior art, Chinese patent literature CN105072049A, open (bulletin) day 2015.11.18, the resource allocation methods and device of a kind of multi-level flexible application in data-oriented center are disclosed, multilayer is belonged to Secondary cloud application technical field.The technical method obtains the multi-level application request of user's proposition first;Secondly to the bandwidth of each layer Demand carries out descending arrangement;It next is each Layer assignment resources of virtual machine according to the sequence of the bandwidth demand of each layer from high to low, Specific as follows: the virtual machine quantity needed first to this layer request is calculated according to the tree topology under current cloud platform to be owned The feasible vector FVl of allocation plan, secondly, obtaining each of the links to the optimal distributing scheme of this layer by recursion method;It connects down It calculates the quantity of " on demand operation " virtual resource, and is reserved in the data center;The user is finally exported to answer at many levels With the optimal distributing scheme of request.But the technology can only solve the resource allocation problem of tree-like typical data center network, can not Solve the VDC imbedding problem of general topology data center.Exclusive path is only existed between tree topology two-server, routing is asked It inscribes very simple;And there is mulitpath mostly between two-server in typical data center network at present, with tree topology In there were significant differences.

Chinese patent literature CN105103506A, open (bulletin) day 2015.11.25 disclose a kind of for being cloud The method and system of the non-homogeneous bandwidth request allocation bandwidth in network is calculated, wherein virtual network includes one or more virtual The first set of interchanger, one or more physics clothes of one or more of virtual switch management hosts virtual machine (VM) The second set of business device.Method starts from the request that more than first a VM are received by a virtual switch, wherein a VM more than first In at least one VM contain the bandwidth different from the bandwidth of remaining VM in one or more VM.Then it is handed over by calculating with virtual It changes planes the set of associated range of distribution (allocation-range, AR), it is determined whether receive the request of more than first a VM, Each AR in middle AR set indicates at least one discontinuous VM allocation space in virtual switch, then for the request point With VM.The algorithm that the technology is related to is known as range of distribution algorithm.But the technology can only solve tree-like typical data center network VDC imbedding problem can not solve the VDC imbedding problem of general topology data center.

Summary of the invention

The present invention In view of the above shortcomings of the prior art, proposes that a kind of non-homogeneous bandwidth evaded based on congestion is virtual Data center is embedded in implementation method, solves putting for the routing issue and virtual machine in virtual data center (VDC) imbedding problem Problem is set, VDC insertion success rate more higher than the prior art can be obtained.

The invention is realized by the following technical scheme:

The present invention relates to a kind of non-homogeneous bandwidth VDC evaded based on congestion to be embedded in implementation method, and VM is pressed bandwidth demand With the sequence sequence successively decreased, first it is sequentially placed in server with adaptation search method for the first time;When for the first time adaptation search method without Method starts perturbation procedures when placing the VM, i.e., be targeting with the most congestion link of physical network, this link load is contributed in search Maximum bottleneck server preferentially will re-start the sequence after the smallest VM unloading of bandwidth required in bottleneck server and put It sets.

The most congestion link has the link of maximum link utilization that is, in physical network, which is expressed as:Wherein maximum link utilization is expressed asE indicates that physical link, E indicate Physical link set, ueIndicate the maximum load of physical link e, CeThe remaining bandwidth of physical link e, specifically uses linear gauge It draws optimal method for routing or K-widest path load route equalization method is calculated:

A) when using the optimal method for routing of linear programming, the maximum link utilization μ is to solve for following linear gauge The target value drawn:

Minimizeμ

Subject to:

Flow conservation constraints:

Link constraint, i.e., the ratio between the load of each link and remaining bandwidth are no more than maximum link utilization:

Wherein: routing variableValue range are as follows:

The constraint of dual variable sum are as follows:

B) when using K-widest path load route equalization method, then linear programming is calculated first and obtains physical link e Maximum load, as ue, then all by findingIn maximum value obtain maximum link utilization μ, calculate maximum load ueLinear programming specifically:

Subject to

Wherein: s and d indicates server, and Q is the set for being at least assigned with the server of a virtual machine,For from service Device s to server d passes through the route assignment variable of link e, is determined by being responsible for proportional routing algorithm,For linear programming Dual variable.

The link load contribution, passes through congestion coefficient f[s] is indicated,Wherein: | Vs| it is the quantity of physical server, most congestion link, r indicates non-homogeneous bandwidth VDC request, Πr=π (i, j) | i=1 ..., N;J=1 ..., | Vs| be The virtual machine of r is requested to place combination, in which: to place variable π (i, j)=1 when VM i is placed on server j;Otherwise π (i, j) =0, N are the virtual machine quantity for requesting r), μ (Πr) it is to place combination ΠrCorresponding peak use rate, in the meter of congestion coefficient During calculation, consider that all interim placements for causing network congestion are combined, i.e. { Πr|μ(Πr)>1}。

The adaptation search method for the first time, specifically includes the following steps:

The maximum virtual machine of bandwidth is selected in step 1, the virtual machine set X never placed.When VM i is selected, by the void Quasi- machine, which is placed into candidate collection S [i] first, not will lead to the server of network congestion, and adaptation search for the first time can skip taboo Server in table Tabu [i].

Step 2, when VM i is temporarily placed into server j, using maximum link utilization μ measure physical network congestion Degree.Once detecting μ > 1, illustrate that VM i, which is placed into server j, can generate network congestion, it is necessary to cancel this and invalid put Combination is set, continues to attempt to VM i being placed into next server.

Step 3, when any link of physical network does not all block, then return to step 1 and continue to place next void Quasi- machine, until all virtual machines are all successfully placed

The perturbation procedures pass through most congestion link firstFind the bottle that most flows are sent to most congestion link Neck serverI.e. bottleneck server byIt is calculated;Then from the serverMiddle removal is minimum BandwidthAnd by bottleneck serverCongestion coefficientZero is reset to, i.e. the perturbation priority of the server is reduced to most It is low.

It recycles in order to prevent, the present invention is by serverIt is added toIntroduce taboo listPlaced subsequent Forbid the virtual machine that will have been unloaded in journeyDuplicate allocation is into bottleneck server

Technical effect

Compared with prior art, the present invention has the characteristics that congestion aware, if it find that since network congestion can not place When one VM, by some the allocated virtual machines of migration of selectivity, help the flow for discharging heavy congestion chain road negative It carries.Since insertion success rate and the income of infrastructure provider are directly linked, the performance of present invention insertion success rate balancing method. Simulation result confirms: the insertion success rate of congestion bypassing method is significantly higher than for the first time very close to the retrogressive method of exponential time Three kinds of methods such as adaptation, adjacent adaptation and greedy method.In the special case of tree-shaped physical topology, emulation knot according to the present invention The embedded performance of fruit, congestion bypassing method is better than existing range of distribution method.Mentioned method, which is equally applicable to uniform bandwidth, asks It asks, according to the optimal routing of linear programming for solution, the insertion success rate of congestion bypassing method is apparently higher than HVC-ACE method.

Detailed description of the invention

Fig. 1 is the method for the present invention schematic diagram;

In figure: a is free physical link bandwidth;B is third time iteration schematic diagram;C is that the failure of adaptation method for the first time swashs Perturbation procedures living;D is final scheme;

Fig. 2 evades algorithm (congestion-aware) for congestion and is adapted to backtracking algorithm (backtracking), for the first time (first-fit), the property of adjacent adaptation (next-fit) and greedy algorithm (greedy) in Fat-tree data center network Can compare: abscissa is virtual machine number, and ordinate is insertion success rate;

Fig. 3 be congestion evade algorithm with backtracking algorithm, for the first time be adapted to, it is adjacent be adapted to and greedy algorithm in VL2 data center Performance in network compares: abscissa is virtual machine number, and ordinate is insertion success rate;

Fig. 4 be congestion evade algorithm with backtracking algorithm, for the first time be adapted to, it is adjacent be adapted to and greedy algorithm in BCube data Performance in heart network compares: abscissa is virtual machine number, and ordinate is insertion success rate;

Fig. 5 be congestion evade algorithm with recall algorithm, be adapted to, adjacent be adapted to and fortune of the greedy algorithm in BCube for the first time The row time compares: abscissa is virtual machine number, and ordinate is runing time (unit: second);

Fig. 6 be congestion evade algorithm with recall algorithm, for the first time be adapted to, range of distribution algorithm (allocation-range) exists The performance of tree data central site network compares: abscissa is virtual machine number, and ordinate is insertion success rate;

Fig. 7 is the performance pair that congestion evades algorithm and the uniform bandwidth VDC request of HVC-ACE algorithm process in Fat-tree Than:

Abscissa is virtual machine number, and ordinate is insertion success rate;

Fig. 8 is the performance comparison that congestion evades algorithm and the uniform bandwidth VDC request of HVC-ACE algorithm process in BCube: horizontal Coordinate is virtual machine number, and ordinate is insertion success rate.

Specific embodiment

Assuming that V indicates the set (V of node in a physics DCNsIndicate server set, V-VsIndicate interchanger collection Close), E indicates physics link set.|Vs| a server with 1 ..., | Vs| mark, and interchanger with | Vs|+1 ..., | V | label.The set of the input/output link at the both ends node v (v ∈ V) uses E respectively+(v) and E-(v) it indicates.Assuming that server j∈VsOn have ajA Free Slots accommodate new virtual machine, the corresponding virtual machine of a slot.Definition e ∈ E is physical link.

According to Hose model, non-homogeneous bandwidth VDC request r can use vector (N, B1, B2..., BN) abstractdesription, in which: N indicates the quantity of virtual machine, B1, B2..., BNRespectively indicate the bandwidth demand of N number of virtual machine.

Define Πr=π (i, j) | i=1 ..., N;J=1 ..., | Vs| it is that the virtual machine of r is requested to place combination, in which: It places variable π (i, j)=1: as virtual machine i (bandwidth Bi) it is placed on server j;Otherwise, π (i, j)=0.

Defining Q is the set for being at least assigned with the server of a virtual machine, it is clear that | Q |≤min (| Vs|, N).

It defines reception/transmission that b (s) is server s ∈ Q and converges flow, it is known that, b (s) is limited on this server s The total bandwidth of all virtual machines is not as follows on server s with other for the total bandwidth of all virtual machines:

Specific step is as follows for the present embodiment:

Step 1, initialization of variable, comprising: the virtual machine VM set that do not place: X={ VM1, VM2 ... VMN };VM i's Avoid server setThe congestion factor f of server[s] ← 0, s=1 ... | Vs|;The number of iterations H←0;

If step 2, the VM not placed gatherIt then returns and is embedded in successfully.If having reached maximum number of iterations H=| Vs| N then returns to insertion failure.Otherwise the maximum virtual machine of bandwidth demand is selected from X, might as well be set as VM i, and add up the number of iterations H ←H+1.The standby server set S [i] of VM i is calculated, server j belongs to the condition of the candidate server set S [i] of VM i Are as follows: it has a vacant slot, and inlet/outlet total surplus bandwidth is not less than the virtual machine for being assigned to server j and other virtual Message capacity between machine.If S [i] Tabu [i] be empty set, return insertion failure.

Step 3, first with adaptation search method for the first time attempt for VM i to be placed into set S [i] in Tabu [i] some properly Server in.Assuming that by the experimental merging server j of VM i, and calculate physical network maximum link utilization μ.If μ > 1 Prove that there are congestions in network, it is necessary to cancel this invalid placement group and merge the congestion factor f for updating server[s] continues It attempts VM i being placed into next server.If μ≤1 proves that, without congestion in network, VM i is placed in server j: π by confirmation (i, j) ← 1;And by VM i never place set in remove: X ← X { VMi }, return to step 2 and continue to place next virtual machine.

Step 4 triggers following perturbation procedures if above-mentioned adaptation search method failure (failing correct placement VMi) for the first time:

It finds and maximum bottleneck server is contributed to network congestionSo thatBy minimum bandwidth void Quasi- machine (is denoted as) from bottleneck serverMiddle removal,It will be unloadedIt places back in and does not place collection In conjunctionReset congestion factor:And update taboo list: Step 2 is then return to continue to place next virtual machine.

The above placement process calculates maximum link utilization μ, needs clear route assignment.Linear gauge is used in the present embodiment It draws optimal method for routing or K-widest path load route equalization method is calculated, in which: the optimal routing of linear programming Method is the energy minimization maximum link utilization in the case where given virtual machine places combination condition, but flow transmits have in a network Any split ratio, multichannel will appear delay variance in path;K-widest path load route equalization method efficiently utilizes Multiple paths of physical network, individual path number K therein are configurable (when K=1 indicate single path routing), are suitable for pair The application of path delay difference sensitivity.

The optimal method for routing of the linear programming specifically refers to: a given virtual machine places combination Πr, Optimization routeIt obtains, optimal route assignment can be obtained by solving linear programming presented below.In order to The congestion in network is avoided, the maximum link utilization that the objective function of linear programming is set as minimizing physical network is denoted as μ.

Minimizeμ

Subject to:

Flow conservation constraints:

Link constraint, i.e., the ratio between the load of each link and remaining bandwidth are no more than maximum link utilization:

Route variableValue range are as follows:

Dual variableWithConstraint are as follows:

Maximum link utilization μ is the target value for being to solve for linear programming.

The K-widest path load route equalization method specifically refers to: first using J.Y.Yen in " Finding The K shortest loopless paths in a network. " (Management Science, vol.17, no.11, Pp.712-716, in 1971) algorithm that proposes, precalculate and store the K of every a pair of of communication serverSPTShortest path.So The present embodiment is from K afterwardsSPTLoop-free shortest path selects K item most broad way diameter, i.e., maximum bottleneck bandwidth.By server s to server The K item of d most broad way diameter is usedIndicate, in which: s, d=1 ... | Vs|, s ≠ d.Definition For server s to the bottleneck bandwidth of server d kth paths, its calculation formula is:

According to the remaining bandwidth capacity of this K paths, the present embodiment is assigned to the uniform flow between server K item Path.The calculation formula of the split ratio of kth paths are as follows:

The calculation formula of route assignment variable on link e are as follows:

When using K-widest path load route equalization method, in the worst case, congestion evades embedded mobile GIS Time complexity be O (N | Vs||E|min(|Vs|, N)3.5L), wherein L is input bit number, | Vs| it is the number of physical server Amount, N is the quantity of virtual machine in VDC, | E | it is physical link number.When method for routing optimal using linear programming, congestion is evaded The time complexity of embedded mobile GIS be O (N | Vs||E|3.5min(|Vs|, N)7L), it is much higher than load balancing in the worst case Routing.

Congestion is evaded embedded mobile GIS embodiment and is specifically included:

Embodiment as shown in Figure 1 demonstrates congestion and evades insertion in two layers of data center comprising six servers The implementation procedure of algorithm.The remaining slot quantity of six servers is respectively a1=0, a2=2, a3=1, a4=a5=2 and a6=1 With free physical link bandwidth such as Fig. 1 (a) identifies (unit: Mbps).There are three virtual machine, bands for VDC request tool to be embedded Wide demand is respectively 90Mbps, 70Mbps and 60Mbps.

In preceding iteration twice, it is adapted to search method for the first time, VM1 (90Mbps) is placed in server 3 first, then by VM2 (70Mbps) is placed on server 2.However due to network congestion, VM3 (60Mbs) can not be placed into any service by adaptive method for the first time In device.

Such as in third time iteration, if VM2 temporarily to be placed to the 6th server, the present embodiment can find second Congestion occurs for the physical link between interchanger and third server, because its load (82Mbps) has been more than residual capacity (75Mbps), as shown in Fig. 1 (b).The failure of adaptation method activates perturbation procedures for the first time, server 3 can be identified as bottleneck clothes Business device simultaneously therefrom removes VM1, as shown in Fig. 1 (d).In next iteration twice, which can be placed on VM1 new master VM3, is then successfully placed on third server, as shown in Fig. 1 (d) by machine, i.e. second server.

Performance Evaluation: below to algorithm in three kinds of typical data center networks (including Fat-tree, VL2 and BCube) Performance carry out emulation testing, while it is contemplated that tree-like physical network and uniform bandwidth VDC request two kinds of special circumstances, specifically It is as follows:

Three kinds of typical data centers are made of 16 servers in emulation testing, and the rate of physical link is 1Gbps. A physics network is all generated according to a kind of typical data center topology at random in each emulation experiment, it is random to its distribution one The VDC of generation is requested, in which: the bandwidth request of N number of virtual machine, which follows, to be uniformly distributed.Since insertion success rate is directly related to number According to the income of hub facility provider, therefore the present embodiment is mainly using the performance of insertion success rate measure algorithm.In addition to this originally Embodiment also pays close attention to the runing time of algorithm, because this is closely related with the feasibility and user experience of algorithm.

The present embodiment and backtracking algorithm, for the first time adaptation algorithm and adjacent adaptation algorithm and greedy in multi-path data central site network The performance of greedy algorithm compares:

Assuming that this five kinds of algorithms are all made of the routing of K most broad way diameter load balancing, individual path number K=2.

In the test, server and physical link have full capacity with Probability p, and server residue slot count is with probability 1- P obeys being uniformly distributed between [0,4], and the remaining bandwidth of physical link obeys uniformly dividing between [0,1Gbps] with probability 1-p Cloth.The bandwidth B of VDC requesti(j=1 ..., N) follows being uniformly distributed between [100Mbps, 700Mbps].

Fig. 2~Fig. 4 shows that average success rate increases the simulation result of (N is from 2 to 10) with virtual machine number, in which: Probability p =0.5.As the success rate highest of expected backtracking algorithm, the reason is that: exclusive segment not will lead to an effective solution scheme Placement combination it is outer, all possible placements of backtracking algorithm search is combined, and cost is the runing time complexity of exponential increase.It is imitative True result be also shown that congestion evade algorithm performance it is very close recall algorithm, and be adapted to than for the first time, adjacent adaptation and greediness Algorithm has significant advantage.

As shown in figure 5, comparing the runing time of the lower five kinds of algorithms of BCube network.The runing time of backtracking algorithm is higher than It is more than two orders of magnitude of other heuritic approaches.Meanwhile congestion evades the runing time of algorithm than adaptation and adjacent adaptation for the first time It is slightly long, and it is lower than greedy algorithm.Since congestion evades embedded mobile GIS with polynomial time complexity, and recalling algorithm has finger Number time complexity, it is contemplated that as virtual machine number/server count etc. increases, the congestion that the present embodiment proposes evades algorithm and exists Advantage in terms of runing time can rapid expansion.To sum up apparently, congestion evade algorithm provided for VDC imbedding problem it is multiple in the time The solution of good tradeoff between miscellaneous degree and performance.

The present embodiment is compared with the prior art (range of distribution algorithm) in tree data central site network:

Emulation uses three layers of tree network comprising 20 servers, and core switch connects two convergence switches, Every convergence switch is connected to two frame topcross.Five servers are connected to a frame topcross.Service-Port Rate is 1Gbps, and the link bandwidth between convergence switch and frame topcross is 5Gbps.The present embodiment passes through emulation ratio Evade algorithm and backtracking algorithm compared with mentioned congestion, is adapted to for the first time flat with range of distribution algorithm (allocation-range) algorithm It is embedded in success rate, as shown in Figure 6.Backtracking algorithm can obtain highest insertion success rate, but regrettably it in tree topology Under still have exponential time complexity.Congestion evade the very close backtracking algorithm of success rate of embedded mobile GIS as a result, and bright It is aobvious to be higher than other two kinds of algorithms.

Uniform bandwidth VDC the present embodiment and HVC- algorithm comparison:

As uniform bandwidth VDC, all N number of virtual machine bandwidth demands are equal.Congestion evades algorithm using linear gauge It draws and obtains optimal path, optimization aim is to minimize network congestion (maximum link utilization).Fig. 7 and 8 is respectively in Fat-tree In BCube, the performance that congestion evades insertion Yu HVC-ACE embedded mobile GIS is compared, congestion evades algorithm ratio HVC-ACE success Rate is significantly increased.

Above-mentioned specific implementation can by those skilled in the art under the premise of without departing substantially from the principle of the invention and objective with difference Mode carry out local directed complete set to it, protection scope of the present invention is subject to claims and not by above-mentioned specific implementation institute Limit, each implementation within its scope is by the constraint of the present invention.

Claims (6)

1. it is a kind of evaded based on congestion non-homogeneous bandwidth VDC insertion implementation method, which is characterized in that by VM by bandwidth demand with The sequence sequence successively decreased first is sequentially placed in server with adaptation search method for the first time;When for the first time, adaptation search method can not Start perturbation procedures when placing the VM, i.e., is targeting with the most congestion link of physical network, this link load is contributed most in search Big bottleneck server preferentially will re-start the sequence after the smallest VM unloading of bandwidth required in bottleneck server and put It sets;
The most congestion link has the link of maximum link utilization that is, in physical network, which is expressed as:Wherein maximum link utilization is expressed asE indicates that physical link, E indicate Physical link set, ueIndicate the maximum load of physical link e, CeIndicate the remaining bandwidth of physical link e, Πr=π (i, j) | I=1 ..., N;J=1 ..., | Vs| it is that non-homogeneous bandwidth VDC requests the virtual machine of r to place combination, in which: | Vs| it is physics clothes The quantity of business device, π (i, j) are to place variable, and N is the quantity of virtual machine in VDC.
2. according to the method described in claim 1, it is characterized in that, the maximum link utilization, specifically use linear programming Optimal routing algorithm or Load Balance Routing Algorithms are calculated:
A) when using the optimal routing algorithm of linear programming, the maximum link utilization μ is to solve for following linear programming Target value:
Minimizeμ
Subject to:
Flow conservation constraints:
Link constraint, i.e., the ratio between the load of each link and remaining bandwidth are no more than maximum link utilization:
Wherein: passing through the route assignment variable of link e from server s to server dValue range are as follows:
Dual variableWithConstraint are as follows:
B) when using Load Balance Routing Algorithms, then linear programming is calculated first and obtains the maximum load of physical link e, as ue, Then all by findingIn maximum value obtain maximum link utilization μ, calculate maximum load ueLinear programming it is specific Are as follows:
Subject to
Wherein: s and d indicates that server, Q are the set for being at least assigned with the server of a virtual machine, and b (s) is server s ∈ Flow, route assignment variable are converged in the reception of Q/transmissionIt is determined by being responsible for proportional routing algorithm,For linear programming Dual variable.
3. according to the method described in claim 1, it is characterized in that, the described link load contribution passes through congestion coefficient fΣ[s] table Show,Wherein: s and d indicates server, and Q is at least It is assigned with the set of the server of a virtual machine, | Vs| it is the quantity of physical server,It is most congestion link, r is indicated Non-homogeneous bandwidth VDC request, Πr=π (i, j) | i=1 ..., N;J=1 ..., | Vs| it is that the virtual machine of r is requested to place combination, Wherein: placing variable π (i, j)=1 when VM i is placed on server j;Otherwise π (i, j)=0, N is the virtual machine number for requesting r Amount, μ (Πr) it is to place combination ΠrCorresponding peak use rate, in the calculating process of congestion coefficient, consideration is all to cause net The interim placement of network congestion is combined, i.e. { Πr|μ(Πr)>1}。
4. according to the method described in claim 1, it is characterized in that, the adaptation search method for the first time, specifically includes the following steps:
The maximum virtual machine of bandwidth is selected in step 1, the virtual machine set X never placed;When VM i is selected, by the virtual machine Being placed into candidate collection S [i] first not will lead to the server of network congestion, and adaptation search for the first time can skip taboo list Server in Tabu [i];
Step 2, when VM i is temporarily placed into server j, using maximum link utilization μ measure physical network congestion journey Degree;Once detecting μ > 1, illustrate that VM i, which is placed into server j, can generate network congestion, it is necessary to cancel this invalid placement Combination, continues to attempt to VM i being placed into next server;
If any link of step 3, physical network does not all block, returns to step 1 and continues to place next virtual machine, Until all virtual machines are all successfully placed.
5. according to the method described in claim 1, it is characterized in that, the perturbation procedures, first pass through most congestion link Find the bottleneck server that most flows are sent to most congestion linkI.e. bottleneck server byMeter It obtains;Then from the bottleneck serverMiddle removal lowest-bandwidthAnd by bottleneck serverCongestion coefficientWeight It is set to zero, i.e. the perturbation priority of the bottleneck server is reduced to minimum, in which: s indicates server, and Q is at least to be assigned with one The set of the server of virtual machine.
6. according to the method described in claim 5, it is characterized in that, recycle in order to prevent, by bottleneck serverIt is added to's Introduce taboo listForbid the virtual machine that will have been unloaded in subsequent placement processDuplicate allocation is to bottleneck server In
CN201510932813.9A 2015-12-15 2015-12-15 Implementation method is embedded in based on the non-homogeneous bandwidth virtual data center that congestion is evaded CN105490959B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510932813.9A CN105490959B (en) 2015-12-15 2015-12-15 Implementation method is embedded in based on the non-homogeneous bandwidth virtual data center that congestion is evaded

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510932813.9A CN105490959B (en) 2015-12-15 2015-12-15 Implementation method is embedded in based on the non-homogeneous bandwidth virtual data center that congestion is evaded

Publications (2)

Publication Number Publication Date
CN105490959A CN105490959A (en) 2016-04-13
CN105490959B true CN105490959B (en) 2019-04-05

Family

ID=55677685

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510932813.9A CN105490959B (en) 2015-12-15 2015-12-15 Implementation method is embedded in based on the non-homogeneous bandwidth virtual data center that congestion is evaded

Country Status (1)

Country Link
CN (1) CN105490959B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106535242A (en) * 2016-09-30 2017-03-22 东南大学 Wireless cloud computing system performance prediction method

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8027354B1 (en) * 2009-04-29 2011-09-27 Cisco Technology, Inc. Network consolidation for virtualized servers
CN102981888A (en) * 2012-10-16 2013-03-20 北京华胜天成科技股份有限公司 Virtualization implementing method for Power server
CN103797463A (en) * 2011-07-27 2014-05-14 阿尔卡特朗讯公司 Method and apparatus for assignment of virtual resources within a cloud environment
CN105103506A (en) * 2013-03-27 2015-11-25 爱立信股份有限公司 Network function virtualization method and device

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130268672A1 (en) * 2012-04-05 2013-10-10 Valerie D. Justafort Multi-Objective Virtual Machine Placement Method and Apparatus

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8027354B1 (en) * 2009-04-29 2011-09-27 Cisco Technology, Inc. Network consolidation for virtualized servers
CN103797463A (en) * 2011-07-27 2014-05-14 阿尔卡特朗讯公司 Method and apparatus for assignment of virtual resources within a cloud environment
CN102981888A (en) * 2012-10-16 2013-03-20 北京华胜天成科技股份有限公司 Virtualization implementing method for Power server
CN105103506A (en) * 2013-03-27 2015-11-25 爱立信股份有限公司 Network function virtualization method and device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
A Perturbation Algorithm for Embedding Virtual Data Centers in Multipath Networks;Jian Zou等;《Globecom 2014 - Symposium on Selected Areas in Communications: GC14 SAC Cloud Networks》;20150212;正文第2240-2245页

Also Published As

Publication number Publication date
CN105490959A (en) 2016-04-13

Similar Documents

Publication Publication Date Title
CN102112981B (en) Distribution of virtual machines in a communication network
US9497112B2 (en) Virtual data center allocation with bandwidth guarantees
US9065734B2 (en) Network bandwidth allocation in multi-tenancy cloud computing networks
Gan et al. Genetic simulated annealing algorithm for task scheduling based on cloud computing environment
Alicherry et al. Network aware resource allocation in distributed clouds
US8862744B2 (en) Optimizing traffic load in a communications network
Zheng et al. An approach for cloud resource scheduling based on Parallel Genetic Algorithm
WO2013163865A1 (en) Virtual machine hot migration and deployment method, server and cluster system
Marquezan et al. Distributed autonomic resource management for network virtualization
CN102082692A (en) Method and equipment for migrating virtual machines based on network data flow direction, and cluster system
CN102904794A (en) Method and device for mapping virtual network
Fischer et al. ALEVIN-a framework to develop, compare, and analyze virtual network embedding algorithms
Rabbani et al. On tackling virtual data center embedding problem
WO2015047833A1 (en) File block placement in a distributed file system network
US9130856B2 (en) Creating multiple NoC layers for isolation or avoiding NoC traffic congestion
CN102223281B (en) Method for remapping resource demand dynamic change based on virtual network
US20150134830A1 (en) Bandwidth Guarantee and Work Conservation
KR101809779B1 (en) Automated traffic engineering for 802.1aq based upon the use of link utilization as feedback into the tie-breaking mechanism
Cohen et al. Almost optimal virtual machine placement for traffic intense data centers
CN102710489B (en) Scheduling system and method for dynamic diversion
Chen et al. Intelligent virtual machine placement for cost efficiency in geo-distributed cloud systems
US9361039B2 (en) Method, related apparatus, and system for virtual network migration
US20170346759A1 (en) Optimizing placement of virtual machines
US9621425B2 (en) Method and system to allocate bandwidth for heterogeneous bandwidth request in cloud computing networks
Hu et al. Resolve the virtual network embedding problem: A column generation approach

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
GR01 Patent grant